HalluciGen-PG

该数据集名为HalluciGen-PG，包含了用于生成释义的示例，其中模型会接收到英文和瑞典语两种可能的给定源句子的释义。每个示例都包括一个源句子、一个正确的假设以及一个包含内在幻觉的错误假设。此外，该数据集还包含了被归类为十一种不同类型错误或添加的幻觉假设，这些错误或添加破坏了蕴涵关系。规模上，数据集包含了138个英文示例和139个瑞典示例，其任务是进行释义生成。

The dataset named HalluciGen-PG contains examples for paraphrase generation, where models are provided with paraphrases of given source sentences in two languages: English and Swedish. Each example consists of a source sentence, a correct hypothesis, and an erroneous hypothesis containing inherent hallucinations. Furthermore, the dataset includes hallucinatory hypotheses categorized into 11 distinct types of errors or additions that break the entailment relationship. In terms of scale, the dataset has 138 English examples and 139 Swedish examples, with the task of the dataset being paraphrase generation.